Out of Range Predicted Recommendations

[gategill]

Describe the bug
I'm on a research project about co-training recommender systems. This involves iteratively training one recommender, adding its predicted ratings into the trainset and training another recommender on that enriched trainset. To add the predicted ratings to the trainset, these ratings should be on the same scale as the original ratings, ie 0-5. However, I noticed that the get_recommendations() method returns ratings that are outside the 0-5 range.

For example, when running sample_main.py, which uses ItemKNN on the movielens_1m dataset, the predictions include results that are beyond the normal 0-5 range. They vary anywhere between 0 and 100 and sometimes go beyond 100

My questions would be:

1. Are those supposed to represent the rating or is it some kind of score?
2. If they are scores and not ratings, how can they be normalised?

To Reproduce

1. Save the recommendations from item_knn_similarity.get_user_recs() in a text file
2. Run sample_main.py
3. Inspect the ratings of recommendations

Expected behaviour
One would expect that all the predicted recommendations would be between 0 and 5.
They vary anywhere between 0 and 100 and sometimes go beyond 100

Screenshots

System details (please complete the following information):

• OS: [Windows 10]
• Python Version [3.9.4]
• Version of the Libraries [0.3.1]